Dietary fat and coronary heart disease: a comparison of approaches for adjusting for total energy intake and modeling repeated dietary measurements

Abstract

Previous cohort studies of fat intake and risk of coronary heart disease (CHD) have been inconsistent, probably due in part to methodological differences and various limitations, including inadequate dietary assessment and incomplete adjustment for total energy intake. The authors analyzed repeated assessment of diet from the Nurses' Health Study to examine the associations between intakes of four major types of fat (saturated, monounsaturated, polyunsaturated, and trans fats) and risk of CHD during 14 years of follow-up (1980-1994) by using alternative methods for energy adjustment. In particular, the authors compared four risk models for energy adjustment: the standard multivariate model, the energy-partition model, the nutrient residual model, and the multivariate nutrient density model. Within each model, the authors compared four different approaches for analyzing repeated dietary measurements: baseline diet only, the most recent diet, and two different algorithms for calculating cumulative average diets. The substantive results were consistent across all models; that is, higher intakes of saturated and trans fats were associated with increased risk of CHD, while higher intakes of monounsaturated and polyunsaturated fats were associated with reduced risk. When nutrients were considered as continuous variables, the four energy-adjustment methods yielded similar associationS. However, the interpretation of the relative risks differed across models. In addition, within each model, the methods using the cumulative averages in general yielded stronger associations than did those using either only baseline diet or the most recent diet. When the nutrients were categorized according to quintiles, the residual and the nutrient density models, which gave similar results, yielded statistically more significant tests for linear trend than did the standard and the partition models.